Lynn A. Streeter

Information Retrieval using a Singular Value Decomposition Model of Latent Semantic Structure

In a new method for automatic indexing and retrieval, implicit higher-order structure in the association of terms with documents is modeled to improve estimates of term-document association, and therefore the detection of relevant documents on the basis of terms found in queries. Singular-value decomposition is used to decompose a large term by document matrix into 50 to 150 orthogonal factors from which the original matrix can be approximated by linear combination; both documents and terms are represented as vectors in a 50- to 150- dimensional space. Queries are represented as pseudo-documents vectors formed from weighted combinations of terms, and documents are ordered by their similarity to the query. Initial tests find this automatic method very promising.

How to tell people where to go: comparing navigational aids

Abstract To compare the effectiveness of navigational aids, drivers attempted to follow routes in unfamiliar environments using either customized route maps, vocal directions, or both. The customized route maps, which included only information relevant to the particular route, were drawn to scale, used colour, included interturn mileages, and showed landmarks. The route to be driven was traced in red. To obtain vocal directions, drivers operated a tape recorder that permitted them to play the next or the previous instruction. Instructions were generated by a set of rules with roughly one set of instructions per turn. Information that was not on the map was not included in the vocal instructions. Drivers who listened to directions drove to destinations in fewer miles, took less time, and showed about 70% fewer errors than the map drivers. The performance of drivers with route maps and voice directions was between that of the map only and voice only drivers.

Acoustic and perceptual indicators of emotional stress.

Tape recordings of telephone conversations of Consolidated Edisons system operator (SO) and his immediate superior (CSO), beginning an hour before the 1977 New York blackout, were analyzed for indications of psychological stress. (SO was responsible for monitoring and switching power loads within the Con Ed network.) Utterances from the two individuals were analyzed to yield several pitch and amplitude statistics. To assess the perceptual correlates of stress, four groups of listeners used a seven-point scale to rate the stress of SO and CSO from either randomized vocal utterances or transcripts of the randomized utterances. Results indicated that whereas CSOs vocal pitch increased significantly with increased situational stress, SOs pitch decreased. Listener ratings of stress from the voice were positively related to average pitch. It appears that listeners stereotype of psychological stress includes elevated pitch and amplitude levels, as well as their increased variability.

Comparing and combining the effectiveness of latent semantic indexing and the ordinary vector space model for information retrieval

Abstract A retrieval system was built to find individuals with appropriate expertise within a large research establishment on the basis of their authored documents. The expert-locating system uses a new method for automatic indexing and retrieval based on singular value decomposition, a matrix decomposition technique related to factor analysis. Organizational groups, represented by the documents they write, and the terms contained in these documents, are fit simultaneously into a 100-dimensional “semantic” space. User queries are positioned in the semantic space, and the most similar groups are returned to the user. Here we compared the standard vector-space model with this new technique and found that combining the two methods improved performance over either alone. We also examined the effects of various experimental variables on the systems retrieval accuracy. In particular, the effects of: term weighting functions in the semantic space construction and in query construction, suffix stripping, and using lexical units larger than a single word were studied.

Applying Speech Synthesis to User Interfaces

Publisher Summary This chapter discusses the various kinds of speech output that are appropriate for an interface. The major advantage of using speech in an interface is its universality; almost everyone understands spoken language. As one of the oldest forms of human communication, humans have evolved to process speech effectively and efficiently. To hear a voice message, one need not be positioned captively as when using a terminal. The listener is also free to use other modalities, such as vision for processing other inputs. Thus, a person can listen to speech for instructions and simultaneously operate a terminal, automobile, or crane. When riding in an automobile or flying an aircraft, one does not want the driver or pilot to alternate between reading and driving. With speech output, there is more likelihood that the message will be attended to. By comparison, written instructions are much easier to ignore. However, speech transmission and storage requires much higher bandwidth then corresponding text.

Use of nonsense‐syllable mimicry in the study of prosodic phenomena

The technique of nonsense‐syllable mimicry of natural utterances has many advantages in the study of prosodic phenomena, especially duration. In analytic studies, the elimination of segmental effects as a factor makes data collection much more efficient, and requires only one segmentation criterion. In perceptual studies, the technique eliminates lexical information without unnatural distortions of the signal. In a series of validation experiments, we have found that (1) the patterns of duration obtained by using this technique were stable and reproducible within and across speakers; and (2) mimicry of different natural models with identical stress patterns and constituent structures produced nearly indistinguishable nonsense‐syllable duration patterns.

Effects of learning English as a second language on the acquisition of a new phonemic contrast

Very sharp discrimination functions for the timing of voice onset relative to stop release characterize perceptual boundaries between certain pairs of stop consonants for adult speakers of many languages. To explore how these discriminations depend on experience, we studied their development among Kikuyu children, whose native language contains no stops in which voicing is substantially delayed relative to stop release (e.g., /p/). Kikuyu distinguishes stops in which voice onset substantially precedes release (prevoiced) from those in which voice onset is nearly simultaneous with release (voiced) for apical and velar places of articulation. However, the language has only a single prevoiced labial stop. Prior to exposure to English, children discriminated prevoiced from voiced labials and voiced from voiceless labials, although these distinctions are not phonemic in Kijuyu. Moreover, the voiced/voiceless discrimination for labials ([ba] versus [pa]) improved markedly with schooling in English, rapidly surp...

Comparing performance of spectral distance measures and neural network methods for vowel recognition

Abstract Neural networks were trained to classify single 20 ms frames of vowels using either perceptually-based spectral representations or LPC spectra as input. Classification performance was compared with performance of several distance measures using nearest-neighbor and mean-distance decision criteria. The non-network distance measures included LPC-residual and cepstral distance measures used in conventional automatic speech recognition systems, as well as a formant-based measure and a new elastic distance measure that explicitly corrects for the effects of spectral tilt. Using an optimal error rate criterion, vowels were discriminated best using the elastic distance measure with the perceptually-based spectrum. Neural networks with LPC spectra as input performed comparably to the better conventional distance measures. While the performance of networks trained with perceptually-based spectral inputs was poorer than that of networks trained with LPC spectra, the features represented by the hidden nodes of this network were more consistent with factors related to human vowel perception.

Acoustic consequences and perceptual indicators of stress

All telephone conversations of Consolidated Edisons system operator beginning an hour before the 1977 New York blackout were studied for indications of emotional stress. The system operator had primary responsibility for monitoring and switching power loads in the New York area. The operators utterances were analyzed in terms of speech rate, utterance length, fundamental frequency, and amplitude parameters as well as utterance type (e.g., questions, orders, and statements). Average fundamental frequency and maximum amplitude reliably decreased with the imminence of the blackout. In a perceptual study all utterances were randomized and presented to listeners, who rated the perceived situational stress of the operator. In contrast to the production data, there were significant positive correlations between perceived stress and: fundamental frequency parameters, maximum amplitude, and time from the blackout. These results are discussed with respect to the type of utterance and the status relationship between the operator and the other party involved in the conversation.

Evaluating spectral distance measures with reference to human perception

Two important criteria for spectral distance measures in automatic speech recognition are: (1) the variance of distances between different tokens of the same utterance should be small and variances of distances between different utterances large, and (2) the pattern of distances across utterances should correlate with perceived phonetic similarity of the utterances. These criteria were used to evaluate several spectral distance measures, including (1) Euclidean distance between two log formant ratios, (2) LP‐residual (“Itakura”) distance, (3) Manhattan distance between linearly spaced points on LP spectra, and (4) Manhattan distance between points on “perceptual” spectra (transforming the frequency scale to barks and convolving with an asymmetric filter of critical bandwidth). Distances were computed among synthetic utterances and among one speakers natural utterances of 11 Dutch vowels produced in isolation. The perceptual similarity data were those reported by Pols, van der Kamp, and Plomp [J. Acoust. ...